Mining data chain graph for fault localization

Fault localization is a challenging task in domainspecific data mining. Most existing works focus on call graph that can find bugs which are associated with control flow. However, there are a lot of bugs related to data flow. In this paper, we presented a data dependency graph in fault localization. The approach at first analyzes the execution of the test suites dynamically, then derives the data dependency graph which reflects data flow traces of any test case. Frequency subgraphs generated which are based on the analysis of these data dependency graph. At last ranking the variables that in those graphs and get the suspicious variables. We have conducted a case study use this approach. The preliminary result shows that our approach is feasible and effective.